init/main.c

   1 /*
   2  *  linux/init/main.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *
   6  *  GK 2/5/95  -  Changed to support mounting root fs via NFS
   7  *  Added initrd & change_root: Werner Almesberger & Hans Lermen, Feb '96
   8  *  Moan early if gcc is old, avoiding bogus kernels - Paul Gortmaker, May '96
   9  *  Simplified starting of init:  Michael A. Griffith <grif@acm.org>
  10  */
  11
  12 #define DEBUG           /* Enable initcall_debug */
  13
  14 #include <linux/types.h>
  15 #include <linux/module.h>
  16 #include <linux/proc_fs.h>
  17 #include <linux/kernel.h>
  18 #include <linux/syscalls.h>
  19 #include <linux/stackprotector.h>
  20 #include <linux/string.h>
  21 #include <linux/ctype.h>
  22 #include <linux/delay.h>
  23 #include <linux/ioport.h>
  24 #include <linux/init.h>
  25 #include <linux/initrd.h>
  26 #include <linux/bootmem.h>
  27 #include <linux/acpi.h>
  28 #include <linux/tty.h>
  29 #include <linux/percpu.h>
  30 #include <linux/kmod.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/kernel_stat.h>
  33 #include <linux/start_kernel.h>
  34 #include <linux/security.h>
  35 #include <linux/smp.h>
  36 #include <linux/profile.h>
  37 #include <linux/rcupdate.h>
  38 #include <linux/moduleparam.h>
  39 #include <linux/kallsyms.h>
  40 #include <linux/writeback.h>
  41 #include <linux/cpu.h>
  42 #include <linux/cpuset.h>
  43 #include <linux/cgroup.h>
  44 #include <linux/efi.h>
  45 #include <linux/tick.h>
  46 #include <linux/interrupt.h>
  47 #include <linux/taskstats_kern.h>
  48 #include <linux/delayacct.h>
  49 #include <linux/unistd.h>
  50 #include <linux/rmap.h>
  51 #include <linux/mempolicy.h>
  52 #include <linux/key.h>
  53 #include <linux/buffer_head.h>
  54 #include <linux/page_cgroup.h>
  55 #include <linux/debug_locks.h>
  56 #include <linux/debugobjects.h>
  57 #include <linux/lockdep.h>
  58 #include <linux/kmemleak.h>
  59 #include <linux/pid_namespace.h>
  60 #include <linux/device.h>
  61 #include <linux/kthread.h>
  62 #include <linux/sched.h>
  63 #include <linux/signal.h>
  64 #include <linux/idr.h>
  65 #include <linux/kgdb.h>
  66 #include <linux/ftrace.h>
  67 #include <linux/async.h>
  68 #include <linux/kmemcheck.h>
  69 #include <linux/sfi.h>
  70 #include <linux/shmem_fs.h>
  71 #include <linux/slab.h>
  72 #include <linux/perf_event.h>
  73 #include <linux/file.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/blkdev.h>
  76 #include <linux/elevator.h>
  77 #include <linux/sched_clock.h>
  78 #include <linux/context_tracking.h>
  79 #include <linux/random.h>
  80
  81 #include <asm/io.h>
  82 #include <asm/bugs.h>
  83 #include <asm/setup.h>
  84 #include <asm/sections.h>
  85 #include <asm/cacheflush.h>
  86
  87 #ifdef CONFIG_X86_LOCAL_APIC
  88 #include <asm/smp.h>
  89 #endif
  90
  91 static int kernel_init(void *);
  92
  93 extern void init_IRQ(void);
  94 extern void fork_init(unsigned long);
  95 extern void mca_init(void);
  96 extern void sbus_init(void);
  97 extern void radix_tree_init(void);
  98 #ifndef CONFIG_DEBUG_RODATA
  99 static inline void mark_rodata_ro(void) { }
 100 #endif
 101
 102 /*
 103  * Debug helper: via this flag we know that we are in 'early bootup code'
 104  * where only the boot processor is running with IRQ disabled.  This means
 105  * two things - IRQ must not be enabled before the flag is cleared and some
 106  * operations which are not allowed with IRQ disabled are allowed while the
 107  * flag is set.
 108  */
 109 bool early_boot_irqs_disabled __read_mostly;
 110
 111 enum system_states system_state __read_mostly;
 112 EXPORT_SYMBOL(system_state);
 113
 114 /*
 115  * Boot command-line arguments
 116  */
 117 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
 118 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
 119
 120 extern void time_init(void);
 121 /* Default late time init is NULL. archs can override this later. */
 122 void (*__initdata late_time_init)(void);
 123
 124 /* Untouched command line saved by arch-specific code. */
 125 char __initdata boot_command_line[COMMAND_LINE_SIZE];
 126 /* Untouched saved command line (eg. for /proc) */
 127 char *saved_command_line;
 128 /* Command line for parameter parsing */
 129 static char *static_command_line;
 130 /* Command line for per-initcall parameter parsing */
 131 static char *initcall_command_line;
 132
 133 static char *execute_command;
 134 static char *ramdisk_execute_command;
 135
 136 /*
 137  * Used to generate warnings if static_key manipulation functions are used
 138  * before jump_label_init is called.
 139  */
 140 bool static_key_initialized __read_mostly = false;
 141 EXPORT_SYMBOL_GPL(static_key_initialized);
 142
 143 /*
 144  * If set, this is an indication to the drivers that reset the underlying
 145  * device before going ahead with the initialization otherwise driver might
 146  * rely on the BIOS and skip the reset operation.
 147  *
 148  * This is useful if kernel is booting in an unreliable environment.
 149  * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
 150  * skipped and devices will be in unknown state.
 151  */
 152 unsigned int reset_devices;
 153 EXPORT_SYMBOL(reset_devices);
 154
 155 static int __init set_reset_devices(char *str)
 156 {
 157         reset_devices = 1;
 158         return 1;
 159 }
 160
 161 __setup("reset_devices", set_reset_devices);
 162
 163 static const char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
 164 const char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
 165 static const char *panic_later, *panic_param;
 166
 167 extern const struct obs_kernel_param __setup_start[], __setup_end[];
 168
 169 static int __init obsolete_checksetup(char *line)
 170 {
 171         const struct obs_kernel_param *p;
 172         int had_early_param = 0;
 173
 174         p = __setup_start;
 175         do {
 176                 int n = strlen(p->str);
 177                 if (parameqn(line, p->str, n)) {
 178                         if (p->early) {
 179                                 /* Already done in parse_early_param?
 180                                  * (Needs exact match on param part).
 181                                  * Keep iterating, as we can have early
 182                                  * params and __setups of same names 8( */
 183                                 if (line[n] == '\0' || line[n] == '=')
 184                                         had_early_param = 1;
 185                         } else if (!p->setup_func) {
 186                                 pr_warn("Parameter %s is obsolete, ignored\n",
 187                                         p->str);
 188                                 return 1;
 189                         } else if (p->setup_func(line + n))
 190                                 return 1;
 191                 }
 192                 p++;
 193         } while (p < __setup_end);
 194
 195         return had_early_param;
 196 }
 197
 198 /*
 199  * This should be approx 2 Bo*oMips to start (note initial shift), and will
 200  * still work even if initially too large, it will just take slightly longer
 201  */
 202 unsigned long loops_per_jiffy = (1<<12);
 203
 204 EXPORT_SYMBOL(loops_per_jiffy);
 205
 206 static int __init debug_kernel(char *str)
 207 {
 208         console_loglevel = 10;
 209         return 0;
 210 }
 211
 212 static int __init quiet_kernel(char *str)
 213 {
 214         console_loglevel = 4;
 215         return 0;
 216 }
 217
 218 early_param("debug", debug_kernel);
 219 early_param("quiet", quiet_kernel);
 220
 221 static int __init loglevel(char *str)
 222 {
 223         int newlevel;
 224
 225         /*
 226          * Only update loglevel value when a correct setting was passed,
 227          * to prevent blind crashes (when loglevel being set to 0) that
 228          * are quite hard to debug
 229          */
 230         if (get_option(&str, &newlevel)) {
 231                 console_loglevel = newlevel;
 232                 return 0;
 233         }
 234
 235         return -EINVAL;
 236 }
 237
 238 early_param("loglevel", loglevel);
 239
 240 /* Change NUL term back to "=", to make "param" the whole string. */
 241 static int __init repair_env_string(char *param, char *val, const char *unused)
 242 {
 243         if (val) {
 244                 /* param=val or param="val"? */
 245                 if (val == param+strlen(param)+1)
 246                         val[-1] = '=';
 247                 else if (val == param+strlen(param)+2) {
 248                         val[-2] = '=';
 249                         memmove(val-1, val, strlen(val)+1);
 250                         val--;
 251                 } else
 252                         BUG();
 253         }
 254         return 0;
 255 }
 256
 257 /*
 258  * Unknown boot options get handed to init, unless they look like
 259  * unused parameters (modprobe will find them in /proc/cmdline).
 260  */
 261 static int __init unknown_bootoption(char *param, char *val, const char *unused)
 262 {
 263         repair_env_string(param, val, unused);
 264
 265         /* Handle obsolete-style parameters */
 266         if (obsolete_checksetup(param))
 267                 return 0;
 268
 269         /* Unused module parameter. */
 270         if (strchr(param, '.') && (!val || strchr(param, '.') < val))
 271                 return 0;
 272
 273         if (panic_later)
 274                 return 0;
 275
 276         if (val) {
 277                 /* Environment option */
 278                 unsigned int i;
 279                 for (i = 0; envp_init[i]; i++) {
 280                         if (i == MAX_INIT_ENVS) {
 281                                 panic_later = "env";
 282                                 panic_param = param;
 283                         }
 284                         if (!strncmp(param, envp_init[i], val - param))
 285                                 break;
 286                 }
 287                 envp_init[i] = param;
 288         } else {
 289                 /* Command line option */
 290                 unsigned int i;
 291                 for (i = 0; argv_init[i]; i++) {
 292                         if (i == MAX_INIT_ARGS) {
 293                                 panic_later = "init";
 294                                 panic_param = param;
 295                         }
 296                 }
 297                 argv_init[i] = param;
 298         }
 299         return 0;
 300 }
 301
 302 static int __init init_setup(char *str)
 303 {
 304         unsigned int i;
 305
 306         execute_command = str;
 307         /*
 308          * In case LILO is going to boot us with default command line,
 309          * it prepends "auto" before the whole cmdline which makes
 310          * the shell think it should execute a script with such name.
 311          * So we ignore all arguments entered _before_ init=... [MJ]
 312          */
 313         for (i = 1; i < MAX_INIT_ARGS; i++)
 314                 argv_init[i] = NULL;
 315         return 1;
 316 }
 317 __setup("init=", init_setup);
 318
 319 static int __init rdinit_setup(char *str)
 320 {
 321         unsigned int i;
 322
 323         ramdisk_execute_command = str;
 324         /* See "auto" comment in init_setup */
 325         for (i = 1; i < MAX_INIT_ARGS; i++)
 326                 argv_init[i] = NULL;
 327         return 1;
 328 }
 329 __setup("rdinit=", rdinit_setup);
 330
 331 #ifndef CONFIG_SMP
 332 static const unsigned int setup_max_cpus = NR_CPUS;
 333 #ifdef CONFIG_X86_LOCAL_APIC
 334 static void __init smp_init(void)
 335 {
 336         APIC_init_uniprocessor();
 337 }
 338 #else
 339 #define smp_init()      do { } while (0)
 340 #endif
 341
 342 static inline void setup_nr_cpu_ids(void) { }
 343 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
 344 #endif
 345
 346 /*
 347  * We need to store the untouched command line for future reference.
 348  * We also need to store the touched command line since the parameter
 349  * parsing is performed in place, and we should allow a component to
 350  * store reference of name/value for future reference.
 351  */
 352 static void __init setup_command_line(char *command_line)
 353 {
 354         saved_command_line =
 355                 memblock_virt_alloc(strlen(boot_command_line) + 1, 0);
 356         initcall_command_line =
 357                 memblock_virt_alloc(strlen(boot_command_line) + 1, 0);
 358         static_command_line = memblock_virt_alloc(strlen(command_line) + 1, 0);
 359         strcpy (saved_command_line, boot_command_line);
 360         strcpy (static_command_line, command_line);
 361 }
 362
 363 /*
 364  * We need to finalize in a non-__init function or else race conditions
 365  * between the root thread and the init thread may cause start_kernel to
 366  * be reaped by free_initmem before the root thread has proceeded to
 367  * cpu_idle.
 368  *
 369  * gcc-3.4 accidentally inlines this function, so use noinline.
 370  */
 371
 372 static __initdata DECLARE_COMPLETION(kthreadd_done);
 373
 374 static noinline void __init_refok rest_init(void)
 375 {
 376         int pid;
 377
 378         rcu_scheduler_starting();
 379         /*
 380          * We need to spawn init first so that it obtains pid 1, however
 381          * the init task will end up wanting to create kthreads, which, if
 382          * we schedule it before we create kthreadd, will OOPS.
 383          */
 384         kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
 385         numa_default_policy();
 386         pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
 387         rcu_read_lock();
 388         kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
 389         rcu_read_unlock();
 390         complete(&kthreadd_done);
 391
 392         /*
 393          * The boot idle thread must execute schedule()
 394          * at least once to get things moving:
 395          */
 396         init_idle_bootup_task(current);
 397         schedule_preempt_disabled();
 398         /* Call into cpu_idle with preempt disabled */
 399         cpu_startup_entry(CPUHP_ONLINE);
 400 }
 401
 402 /* Check for early params. */
 403 static int __init do_early_param(char *param, char *val, const char *unused)
 404 {
 405         const struct obs_kernel_param *p;
 406
 407         for (p = __setup_start; p < __setup_end; p++) {
 408                 if ((p->early && parameq(param, p->str)) ||
 409                     (strcmp(param, "console") == 0 &&
 410                      strcmp(p->str, "earlycon") == 0)
 411                 ) {
 412                         if (p->setup_func(val) != 0)
 413                                 pr_warn("Malformed early option '%s'\n", param);
 414                 }
 415         }
 416         /* We accept everything at this stage. */
 417         return 0;
 418 }
 419
 420 void __init parse_early_options(char *cmdline)
 421 {
 422         parse_args("early options", cmdline, NULL, 0, 0, 0, do_early_param);
 423 }
 424
 425 /* Arch code calls this early on, or if not, just before other parsing. */
 426 void __init parse_early_param(void)
 427 {
 428         static __initdata int done = 0;
 429         static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
 430
 431         if (done)
 432                 return;
 433
 434         /* All fall through to do_early_param. */
 435         strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
 436         parse_early_options(tmp_cmdline);
 437         done = 1;
 438 }
 439
 440 /*
 441  *      Activate the first processor.
 442  */
 443
 444 static void __init boot_cpu_init(void)
 445 {
 446         int cpu = smp_processor_id();
 447         /* Mark the boot cpu "present", "online" etc for SMP and UP case */
 448         set_cpu_online(cpu, true);
 449         set_cpu_active(cpu, true);
 450         set_cpu_present(cpu, true);
 451         set_cpu_possible(cpu, true);
 452 }
 453
 454 void __init __weak smp_setup_processor_id(void)
 455 {
 456 }
 457
 458 # if THREAD_SIZE >= PAGE_SIZE
 459 void __init __weak thread_info_cache_init(void)
 460 {
 461 }
 462 #endif
 463
 464 /*
 465  * Set up kernel memory allocators
 466  */
 467 static void __init mm_init(void)
 468 {
 469         /*
 470          * page_cgroup requires contiguous pages,
 471          * bigger than MAX_ORDER unless SPARSEMEM.
 472          */
 473         page_cgroup_init_flatmem();
 474         mem_init();
 475         kmem_cache_init();
 476         percpu_init_late();
 477         pgtable_init();
 478         vmalloc_init();
 479 }
 480
 481 asmlinkage void __init start_kernel(void)
 482 {
 483         char * command_line;
 484         extern const struct kernel_param __start___param[], __stop___param[];
 485
 486         /*
 487          * Need to run as early as possible, to initialize the
 488          * lockdep hash:
 489          */
 490         lockdep_init();
 491         smp_setup_processor_id();
 492         debug_objects_early_init();
 493
 494         /*
 495          * Set up the the initial canary ASAP:
 496          */
 497         boot_init_stack_canary();
 498
 499         cgroup_init_early();
 500
 501         local_irq_disable();
 502         early_boot_irqs_disabled = true;
 503
 504 /*
 505  * Interrupts are still disabled. Do necessary setups, then
 506  * enable them
 507  */
 508         boot_cpu_init();
 509         page_address_init();
 510         pr_notice("%s", linux_banner);
 511         setup_arch(&command_line);
 512         mm_init_owner(&init_mm, &init_task);
 513         mm_init_cpumask(&init_mm);
 514         setup_command_line(command_line);
 515         setup_nr_cpu_ids();
 516         setup_per_cpu_areas();
 517         smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
 518
 519         build_all_zonelists(NULL, NULL);
 520         page_alloc_init();
 521
 522         pr_notice("Kernel command line: %s\n", boot_command_line);
 523         parse_early_param();
 524         parse_args("Booting kernel", static_command_line, __start___param,
 525                    __stop___param - __start___param,
 526                    -1, -1, &unknown_bootoption);
 527
 528         jump_label_init();
 529
 530         /*
 531          * These use large bootmem allocations and must precede
 532          * kmem_cache_init()
 533          */
 534         setup_log_buf(0);
 535         pidhash_init();
 536         vfs_caches_init_early();
 537         sort_main_extable();
 538         trap_init();
 539         mm_init();
 540
 541         /*
 542          * Set up the scheduler prior starting any interrupts (such as the
 543          * timer interrupt). Full topology setup happens at smp_init()
 544          * time - but meanwhile we still have a functioning scheduler.
 545          */
 546         sched_init();
 547         /*
 548          * Disable preemption - early bootup scheduling is extremely
 549          * fragile until we cpu_idle() for the first time.
 550          */
 551         preempt_disable();
 552         if (WARN(!irqs_disabled(), "Interrupts were enabled *very* early, fixing it\n"))
 553                 local_irq_disable();
 554         idr_init_cache();
 555         rcu_init();
 556         tick_nohz_init();
 557         context_tracking_init();
 558         radix_tree_init();
 559         /* init some links before init_ISA_irqs() */
 560         early_irq_init();
 561         init_IRQ();
 562         tick_init();
 563         init_timers();
 564         hrtimers_init();
 565         softirq_init();
 566         acpi_early_init();
 567         timekeeping_init();
 568         time_init();
 569         sched_clock_postinit();
 570         perf_event_init();
 571         profile_init();
 572         call_function_init();
 573         WARN(!irqs_disabled(), "Interrupts were enabled early\n");
 574         early_boot_irqs_disabled = false;
 575         local_irq_enable();
 576
 577         kmem_cache_init_late();
 578
 579         /*
 580          * HACK ALERT! This is early. We're enabling the console before
 581          * we've done PCI setups etc, and console_init() must be aware of
 582          * this. But we do want output early, in case something goes wrong.
 583          */
 584         console_init();
 585         if (panic_later)
 586                 panic("Too many boot %s vars at `%s'", panic_later,
 587                       panic_param);
 588
 589         lockdep_info();
 590
 591         /*
 592          * Need to run this when irqs are enabled, because it wants
 593          * to self-test [hard/soft]-irqs on/off lock inversion bugs
 594          * too:
 595          */
 596         locking_selftest();
 597
 598 #ifdef CONFIG_BLK_DEV_INITRD
 599         if (initrd_start && !initrd_below_start_ok &&
 600             page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
 601                 pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
 602                     page_to_pfn(virt_to_page((void *)initrd_start)),
 603                     min_low_pfn);
 604                 initrd_start = 0;
 605         }
 606 #endif
 607         page_cgroup_init();
 608         debug_objects_mem_init();
 609         kmemleak_init();
 610         setup_per_cpu_pageset();
 611         numa_policy_init();
 612         if (late_time_init)
 613                 late_time_init();
 614         sched_clock_init();
 615         calibrate_delay();
 616         pidmap_init();
 617         anon_vma_init();
 618 #ifdef CONFIG_X86
 619         if (efi_enabled(EFI_RUNTIME_SERVICES))
 620                 efi_enter_virtual_mode();
 621 #endif
 622         thread_info_cache_init();
 623         cred_init();
 624         fork_init(totalram_pages);
 625         proc_caches_init();
 626         buffer_init();
 627         key_init();
 628         security_init();
 629         dbg_late_init();
 630         vfs_caches_init(totalram_pages);
 631         signals_init();
 632         /* rootfs populating might need page-writeback */
 633         page_writeback_init();
 634 #ifdef CONFIG_PROC_FS
 635         proc_root_init();
 636 #endif
 637         cgroup_init();
 638         cpuset_init();
 639         taskstats_init_early();
 640         delayacct_init();
 641
 642         check_bugs();
 643
 644         sfi_init_late();
 645
 646         if (efi_enabled(EFI_RUNTIME_SERVICES)) {
 647                 efi_late_init();
 648                 efi_free_boot_services();
 649         }
 650
 651         ftrace_init();
 652
 653         /* Do the rest non-__init'ed, we're now alive */
 654         rest_init();
 655 }
 656
 657 /* Call all constructor functions linked into the kernel. */
 658 static void __init do_ctors(void)
 659 {
 660 #ifdef CONFIG_CONSTRUCTORS
 661         ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
 662
 663         for (; fn < (ctor_fn_t *) __ctors_end; fn++)
 664                 (*fn)();
 665 #endif
 666 }
 667
 668 bool initcall_debug;
 669 core_param(initcall_debug, initcall_debug, bool, 0644);
 670
 671 static int __init_or_module do_one_initcall_debug(initcall_t fn)
 672 {
 673         ktime_t calltime, delta, rettime;
 674         unsigned long long duration;
 675         int ret;
 676
 677         pr_debug("calling  %pF @ %i\n", fn, task_pid_nr(current));
 678         calltime = ktime_get();
 679         ret = fn();
 680         rettime = ktime_get();
 681         delta = ktime_sub(rettime, calltime);
 682         duration = (unsigned long long) ktime_to_ns(delta) >> 10;
 683         pr_debug("initcall %pF returned %d after %lld usecs\n",
 684                  fn, ret, duration);
 685
 686         return ret;
 687 }
 688
 689 int __init_or_module do_one_initcall(initcall_t fn)
 690 {
 691         int count = preempt_count();
 692         int ret;
 693         char msgbuf[64];
 694
 695         if (initcall_debug)
 696                 ret = do_one_initcall_debug(fn);
 697         else
 698                 ret = fn();
 699
 700         msgbuf[0] = 0;
 701
 702         if (preempt_count() != count) {
 703                 sprintf(msgbuf, "preemption imbalance ");
 704                 preempt_count_set(count);
 705         }
 706         if (irqs_disabled()) {
 707                 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
 708                 local_irq_enable();
 709         }
 710         WARN(msgbuf[0], "initcall %pF returned with %s\n", fn, msgbuf);
 711
 712         return ret;
 713 }
 714
 715
 716 extern initcall_t __initcall_start[];
 717 extern initcall_t __initcall0_start[];
 718 extern initcall_t __initcall1_start[];
 719 extern initcall_t __initcall2_start[];
 720 extern initcall_t __initcall3_start[];
 721 extern initcall_t __initcall4_start[];
 722 extern initcall_t __initcall5_start[];
 723 extern initcall_t __initcall6_start[];
 724 extern initcall_t __initcall7_start[];
 725 extern initcall_t __initcall_end[];
 726
 727 static initcall_t *initcall_levels[] __initdata = {
 728         __initcall0_start,
 729         __initcall1_start,
 730         __initcall2_start,
 731         __initcall3_start,
 732         __initcall4_start,
 733         __initcall5_start,
 734         __initcall6_start,
 735         __initcall7_start,
 736         __initcall_end,
 737 };
 738
 739 /* Keep these in sync with initcalls in include/linux/init.h */
 740 static char *initcall_level_names[] __initdata = {
 741         "early",
 742         "core",
 743         "postcore",
 744         "arch",
 745         "subsys",
 746         "fs",
 747         "device",
 748         "late",
 749 };
 750
 751 static void __init do_initcall_level(int level)
 752 {
 753         extern const struct kernel_param __start___param[], __stop___param[];
 754         initcall_t *fn;
 755
 756         strcpy(initcall_command_line, saved_command_line);
 757         parse_args(initcall_level_names[level],
 758                    initcall_command_line, __start___param,
 759                    __stop___param - __start___param,
 760                    level, level,
 761                    &repair_env_string);
 762
 763         for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
 764                 do_one_initcall(*fn);
 765 }
 766
 767 static void __init do_initcalls(void)
 768 {
 769         int level;
 770
 771         for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++)
 772                 do_initcall_level(level);
 773 }
 774
 775 /*
 776  * Ok, the machine is now initialized. None of the devices
 777  * have been touched yet, but the CPU subsystem is up and
 778  * running, and memory and process management works.
 779  *
 780  * Now we can finally start doing some real work..
 781  */
 782 static void __init do_basic_setup(void)
 783 {
 784         cpuset_init_smp();
 785         usermodehelper_init();
 786         shmem_init();
 787         driver_init();
 788         init_irq_proc();
 789         do_ctors();
 790         usermodehelper_enable();
 791         do_initcalls();
 792         random_int_secret_init();
 793 }
 794
 795 static void __init do_pre_smp_initcalls(void)
 796 {
 797         initcall_t *fn;
 798
 799         for (fn = __initcall_start; fn < __initcall0_start; fn++)
 800                 do_one_initcall(*fn);
 801 }
 802
 803 /*
 804  * This function requests modules which should be loaded by default and is
 805  * called twice right after initrd is mounted and right before init is
 806  * exec'd.  If such modules are on either initrd or rootfs, they will be
 807  * loaded before control is passed to userland.
 808  */
 809 void __init load_default_modules(void)
 810 {
 811         load_default_elevator_module();
 812 }
 813
 814 static int run_init_process(const char *init_filename)
 815 {
 816         argv_init[0] = init_filename;
 817         return do_execve(init_filename,
 818                 (const char __user *const __user *)argv_init,
 819                 (const char __user *const __user *)envp_init);
 820 }
 821
 822 static int try_to_run_init_process(const char *init_filename)
 823 {
 824         int ret;
 825
 826         ret = run_init_process(init_filename);
 827
 828         if (ret && ret != -ENOENT) {
 829                 pr_err("Starting init: %s exists but couldn't execute it (error %d)\n",
 830                        init_filename, ret);
 831         }
 832
 833         return ret;
 834 }
 835
 836 static noinline void __init kernel_init_freeable(void);
 837
 838 static int __ref kernel_init(void *unused)
 839 {
 840         int ret;
 841
 842         kernel_init_freeable();
 843         /* need to finish all async __init code before freeing the memory */
 844         async_synchronize_full();
 845         free_initmem();
 846         mark_rodata_ro();
 847         system_state = SYSTEM_RUNNING;
 848         numa_default_policy();
 849
 850         flush_delayed_fput();
 851
 852         if (ramdisk_execute_command) {
 853                 ret = run_init_process(ramdisk_execute_command);
 854                 if (!ret)
 855                         return 0;
 856                 pr_err("Failed to execute %s (error %d)\n",
 857                        ramdisk_execute_command, ret);
 858         }
 859
 860         /*
 861          * We try each of these until one succeeds.
 862          *
 863          * The Bourne shell can be used instead of init if we are
 864          * trying to recover a really broken machine.
 865          */
 866         if (execute_command) {
 867                 ret = run_init_process(execute_command);
 868                 if (!ret)
 869                         return 0;
 870                 pr_err("Failed to execute %s (error %d).  Attempting defaults...\n",
 871                         execute_command, ret);
 872         }
 873         if (!try_to_run_init_process("/sbin/init") ||
 874             !try_to_run_init_process("/etc/init") ||
 875             !try_to_run_init_process("/bin/init") ||
 876             !try_to_run_init_process("/bin/sh"))
 877                 return 0;
 878
 879         panic("No working init found.  Try passing init= option to kernel. "
 880               "See Linux Documentation/init.txt for guidance.");
 881 }
 882
 883 static noinline void __init kernel_init_freeable(void)
 884 {
 885         /*
 886          * Wait until kthreadd is all set-up.
 887          */
 888         wait_for_completion(&kthreadd_done);
 889
 890         /* Now the scheduler is fully set up and can do blocking allocations */
 891         gfp_allowed_mask = __GFP_BITS_MASK;
 892
 893         /*
 894          * init can allocate pages on any node
 895          */
 896         set_mems_allowed(node_states[N_MEMORY]);
 897         /*
 898          * init can run on any cpu.
 899          */
 900         set_cpus_allowed_ptr(current, cpu_all_mask);
 901
 902         cad_pid = task_pid(current);
 903
 904         smp_prepare_cpus(setup_max_cpus);
 905
 906         do_pre_smp_initcalls();
 907         lockup_detector_init();
 908
 909         smp_init();
 910         sched_init_smp();
 911
 912         do_basic_setup();
 913
 914         /* Open the /dev/console on the rootfs, this should never fail */
 915         if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
 916                 pr_err("Warning: unable to open an initial console.\n");
 917
 918         (void) sys_dup(0);
 919         (void) sys_dup(0);
 920         /*
 921          * check if there is an early userspace init.  If yes, let it do all
 922          * the work
 923          */
 924
 925         if (!ramdisk_execute_command)
 926                 ramdisk_execute_command = "/init";
 927
 928         if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
 929                 ramdisk_execute_command = NULL;
 930                 prepare_namespace();
 931         }
 932
 933         /*
 934          * Ok, we have completed the initial bootup, and
 935          * we're essentially up and running. Get rid of the
 936          * initmem segments and start the user-mode stuff..
 937          */
 938
 939         /* rootfs is available now, try loading default modules */
 940         load_default_modules();
 941 }